Getting started
Install
Section titled “Install”npm i -g @internet4000/feedbai # or without installing: npx @internet4000/feedbai …npm i -g github:internet4000/feedbai # the latest main, unreleasedNode ≥ 22. No dependencies.
1. Run it in your project
Section titled “1. Run it in your project”cd your-projectfeedbaiWith no channels set up yet, feedbai detects the project (git remote, package.json name), looks for channel links already in the repo (matrix.to rooms, bsky.app profiles, GitHub Discussions links), and prints the feedbai add … commands to run. It also prints instructions for an agent, so you can type !feedbai in Claude Code (or ask any agent to run it) and let the agent set it up with you.
2. Add channels
Section titled “2. Add channels”feedbai add matrix '#yourproject:matrix.org'feedbai add bluesky search yourprojectfeedbai add github discussions you/yourprojectadd creates feedbai.config.json at the project root (the git root). Commit it: it holds no accounts or secrets, and every collaborator’s agent then knows where the project’s users talk. Every channel and its options: channels. The file format: configuration.
3. Accounts, if a channel needs one
Section titled “3. Accounts, if a channel needs one”Many channels need no account:
- Public Matrix rooms: invite
@feedbai:matrix.org(the hub’s bot) to the room, and everyone reads it through the hub. - Bluesky search goes through the hub too. Bluesky author feeds and threads are public.
For private rooms, Bluesky mentions, GitHub Discussions, or to read with your own account instead of the hub (on your own, nothing through our servers):
feedbai login matrix --user @you:matrix.org # password login (or --token to paste an access token)feedbai login bluesky --identifier you.bsky.social # an app password: https://bsky.app/settings/app-passwordsfeedbai login github # a token; or set GH_TOKENRun these in your own terminal, not through an agent: they prompt for the secret without echoing it, check it, and save the account to ~/.config/feedbai/config.json and the secret to ~/.config/feedbai/secrets.env (mode 600). A dedicated low-privilege bot account is a good idea. See configuration → accounts.
4. Read feedback
Section titled “4. Read feedback”feedbai # status of each channel + feedback new since your last runfeedbai --since 7d # a time window (doesn't move the "last run" cursor)feedbai --as claude # a separate "last run" cursor per agentBusy channels are summarized per thread; quiet ones cost a few lines. Messages from maintainers are shown first and marked [maintainer]. With a GitHub token (GH_TOKEN or a gh login), bare feedbai also lists the repo’s open issues and pull requests, with their checks and whether they can merge, so new feedback is read next to the work already in progress.
5. Keep listening (optional)
Section titled “5. Keep listening (optional)”feedbai watch --as claude # prints one line per new message; polls less often while quietRun it in the background in your agent session: agents.
6. Turn feedback into issues (optional)
Section titled “6. Turn feedback into issues (optional)”- Hosted hub (invite-only for now: ask in #feedbai): install the feedbai GitHub App on the repo (it can write issues and read
feedbai.config.json, never your code). Messages that mention the bot becomefeedbackissues within about 10 minutes. - Yourself:
feedbai issues --dry-run, thenfeedbai issues. See feedback → issues. - From your CI, with your own bot: no hub, no App. See on your own.
- Concepts: how the pieces fit together
- Agents: the triage workflow (feedback → issues → PRs)
- CLI reference